OpenAI and Anthropic call for binding UK AI safety laws

Representatives from OpenAI and Anthropic urged British lawmakers to replace voluntary pledges with binding statutory frameworks during a Business, Innovation, Science, and Trade Committee hearing. Anthropic head of Europe Liam Booth-Smith and OpenAI vice president Tom Duff Gordon supported mandatory safety rules, including required incident reporting. In contrast, Google DeepMind’s Owen Larter and Meta’s Rob Sherman resisted new legislation, arguing that existing regulators and industry standards are sufficient.

The hearing exposed significant limitations in the current voluntary regime run by the UK AI Safety Institute (AISI). AISI director Henry de Zoete and Cabinet Office civil servant Cat Little confirmed that AISI cannot compel labs to hand over models or prevent dangerous releases. Access for British regulators remains bound to Washington oversight through the US counterpart, CAISI. Booth-Smith noted that Anthropic was initially unable to grant UK AISI full unrestricted access to its Mythos 5.1 model due to US government restrictions, though access was later granted. Whistleblowers including Daniel Kokotajlo, Geoffrey Irving, and Steven Adler—who left OpenAI in late 2024—testified that competitive pressures prevent self-regulation.

The testimony coincided with a warning from MI5 director Ken McCallum regarding potential bioweapon proliferation risks from advanced AI. Committee chair Liam Byrne indicated lawmakers will look toward Britain's upcoming G20 presidency alongside South Korea to push for an international convention. For AI practitioners and deployment teams, this explicit divide signals that statutory compliance, mandatory incident reporting, and pre-deployment safety audits could soon replace informal voluntary frameworks, particularly as cross-border regulatory sharing between bodies like AISI and CAISI becomes formal law.